Introduction

The retrospective analysis of tree growth based on annual rings before, during, and after stress periods is a suitable and abundant approach to assess drought tolerance of trees. Drought stress influenced tree growth at first, even more than physiological processes such as photosynthesis and respiration (McDowell and Sevanto 2010). There is a long tradition in forest science to use diameter or basal area increment at breast height measured on increment cores or stem discs for assessing a tree’s vitality and growth reaction to stress retrospectively. Basal area increment was used as sensitive indicator (Dobbertin 2005) in almost all studies on softwood (Pichler and Oberhuber 2007; Martínez-Vilalta et al. 2012; Zang et al. 2012) as well as on Douglas-fir (Pseudotsuga menziesii [Mirb.] Franco) (Chen et al. 2010; Sergent et al. 2012). Basal area growth at breast height was a sensitive indicator of stress as previous investigations observed a low priority of carbon allocation to stem wood compared to fine roots or branches (Waring and Schlesinger 1985) and narrow and incomplete (Pollanschütz 1975; Elling 1993) or even missing tree rings (Rubner 1910; Elling 1993) in the lower part of the stem under stress. Within a tree, the sensitivity of radial growth to the climate decreased with stem height (Bouriaud et al. 2005). Kramer (1986) and Sterba (1996) showed growth declines mostly in the lower third of the stem during stress periods for softwoods such as Norway spruce (Picea abies (L.) Karst.), Scots pine (Pinus sylvestris L.), and silver fir (Abies alba Mill.). Similar growth patterns were obtained for Norway spruce (Picea abies (L.) Karst.) under drought stress (Sohn et al. 2012) and under ozone stress (Pretzsch et al. 2010).

The behaviour of height growth under drought stress has rarely been investigated on individual mature trees of Scots pine (Mäkinen 1998) and Smith fir (Abies georgei var. smithii) (Wang et al. 2012). This is probably due to the fact that direct and accurate measurement of height increment of mature trees requires felling of trees (Hasenauer and Monserud 1997). Alternatively, young trees or seedlings of Douglas-fir were analysed for which height increment was easily measurable (Newton and Preest 1988; Olszyk et al. 1998). Height growth of most coniferous species is a complex two-year process involving the formation of terminal buds during late summer of the first year and shoot elongation during spring in the second year (Bréda et al. 2006), whereas radial growth was primarily driven by only current growing season precipitation (Feliksik and Wilczyński 2004; Zhang and Hebda 2004; Griesbauer and Green 2010) and more precisely by the occurrence of water limitation (Eilmann et al. 2013; Taeger et al. 2013). Mäkinen (1998) concluded that the predestination of shoot elongation and the weak connection with the current conditions reduced the suitability of height increment for describing the conditions during a specific growing season.

In line with expected changes in the climatic conditions (Schär et al. 2004), Douglas-fir is discussed to be part of forest management strategies in Germany (Spiecker 2010). Under favourable climatic conditions, Douglas-fir growth exceeded that of other softwood species and also under dry conditions indicated a clear advantage over native species such as Scots pine and European larch (Larix decidua Mill.) (Eilmann and Rigling 2012). The reasons are various: Douglas-fir as a softwood has small and sunken stomata, shedding and abundant leaf waxes, and—physiologically—an effective stomatal control mechanism (Lassoie and Salo 1981). Douglas-fir was also most resistant to cavitation as a consequence of drought compared to other Pinaceae (Pinol and Sala 2000).

Forest management must be adapted in order to respond effectively to climate change challenges and mitigation opportunities. In general, site characteristics have a large influence on the occurrence of water stress (Bauwe et al. 2011). Carnwath et al. (2012) considered site condition as an important silvicultural option and showed that Douglas-firs’ basal area was more sensitive to water availability on xeric sites. There is still much debate as to how stand density or individual competitive situations—regulated by thinning or initial spacing—modify the growth reaction patterns of trees in dry years. For instance, thinning enhanced Douglas-fir growth of individual trees as a result of a longer growing period due to the absence of summer drought and higher rates of growth (Aussenac and Granier 1988). For other softwoods, thinning improved the tree radial growth under drought stresses (Laurent et al. 2003; Kohler et al. 2010). By wider spacing, silvicultural practice aims at increasing growth and stability of individual trees. This issue is even more relevant because recent forest management of softwood has moved towards wider initial spacings (Hein et al. 2007). Growing space and rooted soil volume, which have been developed for many years since seedling, seem to foster drought resilience (Sohn et al. 2012).

On the one hand, foresters and forest-yield researchers are interested in volume increment of individual trees and stands, because the economic value depends mainly on volume. On the other hand, radial (or basal area) growth is the preferred and considerable indicator of a change in tree vitality and forest productivity. However, is it acceptable to extrapolate a particular increment loss or acceleration at breast height to the whole stem or even to the whole tree? In order to clarify this, we did not only focus on the diameter growth of individual trees, but we also directly measured the height growth. Hence, we could assess the volume growth under drought stress more precisely. We further analysed which site, tree, and stand characteristics modify the impact of drought on tree increment. By using a framework presented by Lloret et al. (2011), we addressed the following questions:

  1. 1.

    Do height, basal area, and volume growth of Douglas-fir (Pseudotsuga menziesii [Mirb.] Franco) react differently to drought stress?

  2. 2.

    Which characteristics at the individual and at the stand level affect the growth response under drought stress?

Material and methods

Even-aged Douglas-fir (Pseudotsuga menziesii [Mirb.] Franco) thinning trees—selected by a forester—were removed in a thinning intervention in fall/winter 2010/2011. We sampled trees with a diameter at breast height of at least 25 cm with bark, in total 133. In 2003, the trees were about 30 years old. Table 1 gives an overview of the tree and stand describing parameters used.

Table 1 Tree and stand parameters
Full size table

The trees originated from two experimental trials in Bavaria. One site was located at Heigenbrücken (HE) next to the city of Aschaffenburg (North of Bavaria) in the low mountain range Spessart. The other site was located about 15 km south of the city of Ansbach (AN) or 50 km west of the city of Nuremberg. Table 2 summarizes important site characteristics. Both experimental trials were initiated at the beginning of the 1970s in order to comprehend the dependencies of growth on different spacing treatments. In fact, the experimental trials (HE and AN) are divided into smaller single plots at both sites. They are square-shaped with an edge length of 30 m and a buffer zone of 5 m around the plot. The sampled trees represented a wide range of managed Douglas-fir forests, because initial plant density differs within the single plots: 1,000, 2,000, and 4,000 trees per hectare. We did not take into account this kind of information, because preliminary analysis did not show any influence of initial plant density on growth under drought.

Table 2 Site characteristics
Full size table

Figure 1 shows the modified de Martonne aridity index introduced by de Martonne (1926) and applied, for instance, by Uhl et al. (2013). We calculated this index between May and September over the observation period 1998–2008. This index is a ratio of the annual precipitation (in mm) and the sum of annual mean temperature (in °C) and a summand of ten. On both sites, drought occurred in 2003. Drought was more severe at AN.

Fig. 1
figure 1

The de Martonne aridity index—calculated from May to September—characterizes the climate in the growing season over time (de Martonne 1926)

Full size image

Figure 2a, b presents the average monthly temperature for the period from autumn 2002 to autumn 2003 and for a reference period. Similarly, charts were prepared for the monthly precipitation sum separated into the two sites (Fig. 2c, d). A promising approach was introduced by Walter and Lieth (1967) which summarizes temperature and precipitation information and indicates water availability (Fig. 2e, f). Monthly water availability is graphically displayed as precipitation sum (in mm) divided by two, minus average monthly temperature (in °C). Negative values indicate water deficit illustrated by dots, and positive values indicate water surplus illustrated by vertical lines. The dotted area of the Walter and Lieth climate diagram (when the temperature line runs above the precipitation line) clarifies the longer dry periods for AN from June to September of 2003.

Fig. 2
figure 2

Average monthly temperature for Ansbach (a) and Heigenbrücken (b), monthly precipitation sum separated into sites (c, d); within the year 2003, the dotted area is larger for Ansbach (e) than for Heigenbrücken (f)

Full size image

The competitive situation of an individual tree was described by the SDIlocal (Pretzsch and Biber 2010). They defined the competition situation by means of a modified local version of the original stand density index SDI (Reineke 1933). The modified SDIlocal was calculated for a sub-collective of taller trees only in the proximity of the tree of interest, as this area includes its most important competitors. In particular, all neighbouring trees of a tree were considered with a distance of less than twice its mean crown radius. Each tree had on average four competitors. Thinning was performed according to a standardized experimental programme described by Klädtke et al. (2012): when dominant height reached 12 m, a number of 150 future trees was selected. Thinning type was defined as a final crop-tree system. Thinning frequency was determined by the dominant height. A guide curve gave the number of thinning trees per hectare depending on the dominant height. In the past, thinning was done in autumn/winter of 1989, 1994, 1999, and 2004. For this study, the intensity of thinning was quantified by the ratio between SDIlocal before thinning and SDIlocal after thinning, respectively: the higher the ratio, the higher the competition release. Furthermore, we calculated a ratio between height of the particular tree and its five nearest neighbours to explain the social situation (social class) of an individual tree. The crown volume was calculated based on measured crown length (height—crown base height) and crown projection area assuming that the crown shape is simply described by a cone.

Annual increment of height, basal area, and volume

After felling, the current annual increment of height (CAI h ) was measured along the main trunk from the top downwards as the distance between branch whorls to an accuracy of ±2.5 cm. The 133 discs at breast height were polished on a sanding machine using sandpaper with a coarseness of 60–120 grids. Tree-ring width was measured to the nearest 1/100 mm in the four cardinal directions (deviations due to the presence of branches or fissures possible) with a Digital Positiometer (Kutschenreiter and Johann; Digital positiometer, Biritz and Hatzl GmbH, Austria) using the software Lignometer. Crossdating accuracy was checked visually within (four time series per disc) and between ring-width time series. Due to the short period considered, we refrained from using COFECHA (Holmes 1983) as recommended by many long-time investigations. Besides, annual rings of the sampled Douglas-fir stem discs were clearly identifiable. Because of their low percentage of latewood, the annual rings of 2003 were especially obvious. The basal area was calculated by means of quadratic mean radius of the four cardinal directions. The difference between the basal areas of two consecutive years gave the current annual increment of basal area (CAIba). We applied a generalized constant form factor of 0.5 to reduce the volume of a cylinder to tree volume (v = h × d 2/4 × π × 0.5). As the form factor changes very sluggishly with tree size development and our analysis was restricted to a rather short part of tree ontogeny, we assumed it as constant. The difference of two subsequent tree volumes results in the current annual increment of volume (CAI v ).

Height, basal area, and volume increment culminate at different ages and vary in their curve shape. Therefore, we chose the moving average as smoothing, because it was best suited to consider short-term deviations (Riemer 1994; Schlittgen and Streitberg 1997; Pretzsch 2010). Individual series were detrended by Eq. 1:

$$I_{q} = \frac{{{\text{CAI}}_{q} }}{{Y_{q} }}$$
(1)

The parameter I q is the relative growth index of q (height, basal area, or volume). CAI q is the original (measured) increment, and Y q is the increment calculated by moving average value of 5 years (the year itself, two previous years, and two subsequent years). The resulting index curves I h , I ba, and I v were free of age trends. Figure 3 illustrates the method of detrending for the basal area increment of an individual tree.

Fig. 3
figure 3

Original data were smoothed by five-year moving average

Full size image

Lloret’s indices R t and R s for quantifying stress response

For comparing the performance of trees with intra-specific environment under episodic drought stress, we applied the numerical indices for resistance R t and resilience R s introduced and explained in detail by Lloret et al. (2011). R t and R s were calculated individually on the basis of CAI h , CAIba, and CAI v as well as on the basis of the detrended data I h , Iba, and I v for all individual trees. The basic components of the indices were the following:

  • PreDr is the mean mid-term increment in a period of 5 years before the drought period.

  • Dr is the increment in the year of drought 2003 (shoot formation itself is a two-year process—bud formation during late summer of the first year and shoot expansion during spring/early summer in the second year. Describing the height pattern, Dr is the mean increment in the years 2003 and 2004).

  • PostDr is the mean increment in a period of 5 years after the drought period (regarding height only a period of 4 years after the drought period).

  • R t  = Dr/PreDr is the resistance and quantifies the decrease from the pre-drought period (5 years: 1998–2002) to the drought period. The case of R t  = 1 means complete resistance; the more the value falls below 1, the lower the resistance.

  • R s  = PostDr /PreDr is called resilience and represents the ratio between post-drought (5 years: 2004–2008 for basal area and volume, 4 years: 2005–2008 for height) and pre-drought increment. R s  ≥ 1 indicates full recovery or even an increase after the episodic stress, while values of R s  < 1 indicate growth decline and low resilience.

Linear mixed models were performed to verify whether R t or R s were related to any of the following explanatory variables: site, stand density (SDI), intensity of thinning (1999 and 2004), crown volume, competition (SDIlocal), and social class. The models to be fitted were then formulated as

$$\begin{aligned} R_{t} = & \\ a_{0} + & a_{1} {\text{site}} + a_{2} {\text{SDI }} + a_{3} {\text{thinning}}_{1999} + a_{4} {\text{crown volume }} + a_{5} {\text{SDI}}_{\text{local}} + a_{6} {\text{social}}\;{\text{class}} + b_{i} + \varepsilon \\ R_{s} = & \\ a_{0} + & a_{1} {\text{site}} + a_{2} {\text{SDI }} + a_{3} {\text{thinning}}_{1999} + a_{4} {\text{thinning}}_{2004} + a_{5} {\text{crown volume}} + a_{6} {\text{SDI}}_{\text{local}} + a_{7} {\text{social}}\;{\text{class}} + b_{i} + \varepsilon \\ \end{aligned}$$

where i indexed the plot and a 0, a 1, a 2, a 3, a 4, a 5, a 6, and a 7 were the fixed effect coefficients to be estimated. The explanatory variable site (Ansbach, Heigenbrücken) was of categorical type and dummy coded, and the other variables were covariates (quantitative predictors). Additionally, we took into account interactions between site and the remaining explanatory variables.

Beyond these fixed effects, random effects were considered to match the error structure of the data. The parameter b i was a nested random effect at plot level (Pinheiro and Bates 2000; Zuur et al. 2009). The random effects were assumed to be normally distributed with zero mean and constant variance. This nesting-level specific variance was estimated during the model fitting process. The symbol ε represented the independent and identically distributed random error.

The statistical method included three steps: first, the Akaike information criterion (AIC) was used as fitting criterion for selecting the best (multiple) linear model (Burnham and Anderson 2002). This model simplification procedure was carried out to obtain the adequate model. The model with the lowest AIC was selected and used for further analysis. It was also tested whether this model performed better than a model including only random effects. Second, the fixed and random effects of the selected model were estimated, and third, the significance level of each effect was calculated.

All data were analysed using R (R Development Core Team 2012) and the R package lme4 (Bates and Maechler 2009) and languageR (Baayen 2008). The default behaviour of lmer to optimize the maximum likelihood criterion ML was used (Pinheiro and Bates 2000). Normality and homogeneity were checked by visual inspections of diagnostic plots of residuals against fitted values (Zuur et al. 2009, 2010).

Results

The current annual increment of height CAI h , basal area CAIba, and volume CAI v in the period 1998–2008 is shown for both sites HE and AN (Fig. 4a–c). The level of all three kinds of CAI was always higher for HE. The CAI h —averaged per tree and over the reference period of 11 years (1998–2008)—was 0.58 m yr−1 for AN and 0.74 m yr−1 for HE. For the drought year 2003, mean CAI h slumped to 0.37 m yr−1 (AN) and 0.73 m yr−1 (HE). The course of CAIba remained rather constant during the observation period; on average, it was 16.9 cm2 yr−1 for AN and 24.4 cm2 yr−1 for HE. In 2003, CAIba was 11.0 cm2 yr−1 for AN and 20.0 cm2 yr−1 for HE on average over all 133 trees. A similarly synchronous behaviour was observed between the two sites for CAI v . The mean CAI v for AN was 19.9 dm3 yr−1 in 2003 and 30.4 dm3 yr−1 for the 11-year reference period, analogous for HE 40.5 and 46.1 dm3 yr−1.

Fig. 4
figure 4

Current original annual increment of height CAI h (m yr−1) (a), basal area CAIba (cm2 yr−1) (b), and volume CAI v (dm3 yr−1) (c) as well as current detrended annual increment of height I h (d), basal area I ba (e), and volume I v (f) mean with twice standard error bars for 133 Douglas-fir trees from two sites AN (circle) and HE (triangle)

Full size image

The current detrended annual increment of height I h , basal area I ba, and volume I v varied more widely (Fig. 4d–f). Standard deviations of the 11 (mean) annual indices I h , I ba, and I v were higher for AN (0.23, 0.19, and 0.18) than for HE (0.07, 0.09, and 0.06). Looking at the site AN, the high increment of 2002 and the deep descent of the following year 2003 are clearly visible.

Table 3 shows two aspects of tree tolerance to drought-induced low-growth periods: the resistance R t as the growth performance during the drought and the resilience R s as ability to return to the pre-drought growth level. At site AN, growth declined more severely. The height growth was strongly reduced in 2003 and reached only half of the previous period (R t,h , original). In 2003, the basal area increment was 73 % (R t,ba, original) compared to the previous period, and the volume increment was 69 % (R t,v , original). Consequently, volume growth decline (R t,v , original) was 4 % higher than the basal area growth decline (R t,ba, original). Without the age trend, the difference decreased between height, basal area, and volume, but height increment remained most affected by drought (R t,h  < R t,ba, detrended). At site HE, growth in 2003 seemed to be similar to the growth in previous periods (R t , original). The highest growth decline with an average of 16 % was observed for height; for volume, the decline was only 2 %. Volume growth decline of 2003 (R t,v , original) was about 10 % less than the basal area growth decline (R t,ba, original). Considering the age trend, the order between height and basal area changed: R t,ba was most reduced (0.85); R t,h lost only 5 % of increment compared to the height increment of the previous period (detrended data).

Table 3 Resistance R t and resilience R s of height, basal area, and volume growth divided into site
Full size table

With respect to the stress resilience (R s ), the differences between the sites were smaller (Table 3). The mid-term height growth of HE decreased by 27 % (original). Some of this reduction can be attributed to the decreasing age-related curve of the individual current annual height increment CAI h (Fig. 4a), which was confirmed by R s,h (0.76) calculated by the detrended height data Ih. Considering basal area and volume increment, a complete and mid-term recovery was observed (R s,ba and R s,v  > 1, original). Contrary to current annual height growth, the culmination of the individual current annual increment of basal area (CAIba) and of volume (CAI h ) has not been reached. Bearing in mind the age trend, R s,ba and R s,v were close to one (detrended).

The selected models with the lowest AIC values are shown in Table 4 for the original data and in Table 5 for the detrended data, respectively. Predictors with a p value lower than 0.05 are in bold.

Table 4 Original data: resistance, R t , and resilience, R s , as modified by site, tree, and stand characteristics
Full size table
Table 5 Detrended data: resistance, R t , and resilience, R s , as modified by site, tree, and stand characteristics
Full size table

Original data

Primarily, thinning modified growth depression in the year of the drought event and also reduced drought stress at least for some years—whether for height, basal area, or volume growth (Table 4). Furthermore, site and SDI significantly influenced R t,h , and crown volume significantly influenced R s,h and R s,v . We observed a change in significant predictors: Large crown volume influenced the stem growth under drought positively, but did not affect basal area growth. We also scrutinized each model for interactions, because, for instance, due to a storm in winter of 1999, the thinning degree of 2004 was much lower at the site AN (Table 2, stand characteristics for survey 2004). However, by means of a powerful statistical analysis, the large degree of variation within the explanatory variables was considered. No significant interactions were found.

Detrended data

Table 5 shows the results for R t and R s using detrended data. The site gained in importance, as it was significant for R t,h , R t,ba, R t,v , and R s,v . The results showed that the growth of Douglas-fir at the site AN with more severe drought suffered more. Compared to the original data (Table 4), thinning became less important, but had still a positive impact on volume increment (R t,v and R s,v ). On the site of AN (more severe drought), thinning of the year 1999 was twice as important as on the site HE (0.03 + 0.04). The basal area increment of Douglas-fir trees which were thinned in 2004 grew more (R s,ba)—also without the increasing age trend. Simultaneously, recovery of basal area increment (R s,ba) was higher for trees with lower social status. The stand density SDI had a negative effect on R s,h and R s,ba: the higher the SDI, the lower the recovery.

Discussion

The analysis emphasized the relevance and key role of data processing. Annual height, basal area, and volume increment have different growth patterns at the same age. In a first step, we used untransformed data for analysing Rt and Rs as done by Pretzsch et al. (2013) and Taeger et al. (2013), although height, basal area, and volume increment culminate at different ages and vary in their curve shape. So we stayed as close as possible to the originally measured data and avoided nontransparency of the results by trend elimination and indexing. The possible bias by using the original data was assumed to be small due to the short period considered (Pretzsch et al. 2013). In general, R s was assumed to be more influenced than R t , because, in fact, a time interval of 11 years was considered for R s , but only of 6 years for R t . Consequently, the natural curve shape of the current annual increment might become more important for R s : We hypothesized that Douglas-fir height growth course has already culminated and the development of CAI h decreases throughout the period considered (Fig. 4a). On the contrary, CAIba and CAI v increased slightly within the period between 1998 and 2008. In a second step, we kept in mind the still increasing (natural) course of CAIba and CAI v and the already decreasing course of CAI h : we detrended the original data before calculating the Lloret’s indices. The discussion deals with this aspect of data processing.

Stress response of height versus basal area and volume increment

In general, allocation and resulting tree allometry can be modified by stress (Lassoie and Salo 1981). Our analysis revealed different behaviours of height and basal area increment during drought stress. Particularly, during a period of high drought stress like at the site AN, height growth seems to be a more sensitive indicator of vitality than basal area growth—both using original and detrended data. Other studies emphasized the lower sensitivity of the upper part of the trunk to drought based on basal area measurements at different height levels (Sterba 1981; Bouriaud et al. 2005). Our investigation verified those results only for the detrended data for the site HE which was characterized by lower water stress. A possible explanation for the stronger decline of height versus basal area increment might be the occurrence of water stress within the vegetation period. Another explanation might be the restricted water transport to the upper tree parts that reduces growth. In addition, the tree might change its allocation pattern from crown expansion and carbon assimilation—which is mainly achieved by height development—to root extension in order to capture additional water and thus reduce the growth limiting factor (Sharp 2002; Pretzsch et al. 2010). This would be in line with the optimal partitioning theory that a plant always invests in those organs which improve access to the currently limiting resource (McCarthy and Enquist 2007). For instance, light limitation enhanced shoot growth in relation to root growth (Oliver and Larson 1996).

The level of observed basal area growth reduction was consistent with that of Sergent et al. (2012) who found for French Douglas-fir—depending on the site—a reduction of 20–30 % in 2003 compared to the previous years’ growth. However, conclusions extrapolating a particular decline in increment at breast height to the whole stem can result in misunderstandings depending on the site: based on the original data, an overestimation of 4 % of the stem growth decline at the dryer site (AN, R t ) and an underestimation of 12 % of the stem growth decline at the moister site (HE, R t ). Results were also biased, when short-term assessment of growth reactions was only based on original height measurements (R t,h vs. R t,v ): height decline underestimated stem growth decline at both sites by 12 % (HE) and 18 % (AN). When we look at the resilience to drought stress (R s , original data), the average four-year height growth of the subsequent period was 73 and 76 % of the previous period (R s,h ). We found that R s,h underestimated R s,v in a dry stand by 45 %, in a moist stand by 50 %. The basal area defined the volume much more precisely, but—depending on the site—too low (6 %) or too high (11 %). When we transferred the decline in increment at breast height to the whole stem based on detrended data, the situation changed considerably. We observed in all cases for both height and base area increment, deviations from the volume increment equal to or less than 6 %.

Modification of stress response by site, thinning, tree, and stand characteristics

Site, stand density, thinning, individual crown volume, and social class can affect growth response under drought stress (Pretzsch and Dieler 2011). Site played a crucial role in our investigation. A clear relationship existed between growth and soil water deficit for Douglas-fir (Sergent et al. 2012). Precipitation sum appeared to be the most differentiated climatic factor between sites AN and HE and probably was the main reason for the significance of site (Tables 4, 5). Feliksik and Wilczyński (2007) investigated increment cores of different Douglas-fir stands in western and northern Poland and found an interregional variation in tree-ring chronologies mainly determined by air temperature for January–March and by precipitation sum for June–August. Douglas-fir growth seems to be mainly sensitive to the pluvial conditions of summer (June–August). Kantavichai et al. (2010) associated soil moisture content with wood quality. Analysing 76-year-old Douglas-firs from Western Washington, they found that July soil moisture deficit was an important predictor for wood density of annual growth rings. This is very interesting, because many aspects are linked to wood density: on the one hand, wood density was a commonly used predictor and indicator of wood quality (Rais et al. 2014). Although we did not measure early and latewood width separately, the small latewood width in 2003 was very obvious and most helpful for visual cross dating and synchronization. Growth response of Douglas-fir in 2003 was distinct in latewood since rainfall was low in the latewood forming months June–September (Eilmann et al. 2013). On the other hand, wood density was also responsible for the survival of trees during severe droughts. Martinez-Meier et al. (2008) analysed the behaviour of wood density to the 2003 drought by means of about 40-year-old Douglas-firs from provenance trial in France and observed for the surviving trees higher wood density at tree ring and tree level. In our analysis, the low penetrable soil depth for roots (about 30 cm, Table 2) might intensify the drought stress at site AN. A lack of water might occur earlier at AN and might persist even longer (Fig. 1b, c). Up to now, Douglas-fir grew better at HE than at AN, because, generally, water supply was higher due to both the soil conditions as well as the precipitation in the vegetation periods. Consequently, dominant height, total volume production, and level of annual increment were higher (Table 1). During drought, height growth response was influenced most negatively and significantly affected by site (Table 4). For the annual height increment of 1 year, both the weather of the previous year (bud formation) and the current year (shoot expansion) were crucial (Bréda et al. 2006). Douglas-fir is able to form Lamma shoots; this adaption was responsible for up to 20 % of height growth in younger trees (Bongarten 1978). Although we did not separately measure the length of the Lamma shoots, it is very likely that at the site AN, they could not be developed due to the drought stress at summer (Roth and Newton 1996). For AN, a decline in height growth was observed also in the years 2000 and 2001 (Fig. 4a), which might be possible due to the poor water supply. A storm called Lothar caused widespread windthrow in German forests and also at the site AN on 26 December 1999 (Albrecht 2009; Schmidt et al. 2010). As a result of the storm, four single plots could not be used further and had to be dismissed, because most of the trees were windthrown. The trees of the remaining plots at AN were presumably affected by root damages. In the years 2000 and 2001 (Fig. 4a), the damaged trees probably enhanced root growth in relation to shoot growth (Pretzsch et al. 2010).

Our investigation showed that thinning is one of the silvicultural options for adaption and mitigation to climate change. Douglas-fir recovered very well in terms of basal area (R s,ba) and volume (R s,v ) on both sites, and it responded to thinning with more growth due to higher access to light, nutrients, and especially water. This might be due to a combined ecophysiological effect on the water balance which many studies have already analysed before (Donner and Running 1986; Bréda et al. 1995; Laurent et al. 2003): first, soil water availability generally increased in thinned stands. On individual tree level, thinning increased transpiration, but on stand level, transpiration decreased because of the more important effect of reduced stem number. Besides, more rainfall reached the forest soil as a result of reduced canopy water interception (Bréda et al. 1995). For Douglas-fir, the increase of the soil water reserve led to a lower duration and level of water stress in the thinned plot of 19-year-old plantation, especially during the first 3 years after thinning (Aussenac and Granier 1988). For Norway spruce, resilience was more rapid in trees from heavily thinned stands, even if the drought event occurred more than 10 years after the last thinning intervention (Kohler et al. 2010). Second, water-use efficiency may be increased by thinning (Sergent et al. 2012). In contrast, however, Martín-Benito et al. (2010) did not observe any improvement of water-use efficiency for Pinus nigra as a consequence of thinning. In general, Sterba (1981) found that particularly, the basal area increment of the lower part of a tree reacted positively to thinning and not the whole bole. This occurred even years after thinning and may partly explain the high significance of thinning in terms of basal area and volume (standard volume calculation including basal area) (Sterba 1981; Sohn et al. 2012).

The increased portion of water for each tree caused by stand density reduction seems to also be responsible for the relevance of the SDI (Tables 4, 5). The stand density was significant in explaining height growth (original data) and basal area as well as volume growth (detrended data). The stand density index (SDI) influenced growth negatively, i.e. current annual increment of trees from plots with lower tree density was less affected by 2003 drought: It is likely that the increased availability of soil space is associated with a larger root system of the individual tree; water stress occurred later, and growth was less influenced. We barely observed an influence of tree characteristics on growth behaviour around the 2003 drought. A reason for this might be the limited sampling of trees which were regarded as predominant, dominant, and co-dominant [social tree class 1–3 (Kraft 1884)]. Assumably, the effect of competition on the incremental response of Douglas-fir was strongly reduced, but not completely eliminated: large crown volumes improved the growth pattern (original data), and co-dominant trees were less affected by drought than dominant ones (detrended data). New investigations with a more balanced selection of trees have to provide reliable results regarding the influence of tree characteristics on growth performance during drought.

Conclusions

Height growth response to severe drought stress was an even more sensitive indicator than basal area growth. With regard to forest inventories and damage expertise, the short-term assessment of tree’s ability to recover from stress or vitality should ideally respect volume growth. Using original data, we showed that the direct use of just one of these indicators—height or basal area—to estimate relevant volume growth is questioned as it creates errors when the other indicator is neglected at the same time. Growth pattern of basal area described volume growth better than height growth pattern: for the two sites, an underestimation of 6 % and an overestimation of 11 % were obtained. Considering the age trend, both height and basal area growth pattern can be used to indicate volume growth pattern; a false estimation of less than 6 % was observed.

In our investigation, Douglas-fir was characterized by a high drought tolerance, even outside its natural range. Differences in site and site management can significantly influence climate–growth relationships of Douglas-fir. The right choice of site with adequate water supply might be the primary silvicultural option for the adaption of stands to climate change. Secondly, thinning was advantageous for the growth performance, because thinning could decrease the susceptibility of Douglas-fir to drought stress by increasing water availability for the remaining trees. More growing space might lead to shorter stress period for individual trees. In that way, large individual crown volume improved the growth pattern under drought, and large stand density impaired it.

Douglas-fir might be to some extent representative of the behaviour of conifers, and so the results are of general relevance for forest ecology and management. Conifers rather pursue an isohydric stomatal control strategy (for instance, early closure of stomata under moderate droughts) which should prevent hydraulic dysfunction or cavitation (Hartmann 2011). Douglas-fir seems to have ecophysiological adaptations of an isohydric species for avoiding drought stress like in 2003 (Grieu et al. 1988; Pinol and Sala 2000).